Face Generation¶

In this project, you'll use generative adversarial networks to generate new images of faces.

Get the Data¶

You'll be using two datasets in this project:

MNIST
CelebA

Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.

If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Found mnist Data
Found celeba Data

Explore the Data¶

MNIST¶

As you're aware, the MNIST dataset contains images of handwritten digits. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x7f1e4d623668>

CelebA¶

The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x7f1e4d594208>

Preprocess the Data¶

Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.

The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).

Build the Neural Network¶

You'll build the components necessary to build a GANs by implementing the following functions below:

model_inputs
discriminator
generator
model_loss
model_opt
train

Check the Version of TensorFlow and Access to GPU¶

This will check to make sure you have the correct version of TensorFlow and access to a GPU

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.1.0
Default GPU Device: /gpu:0

Input¶

Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:

Real input images placeholder with rank 4 using image_width, image_height, and image_channels.
Z input placeholder with rank 2 using z_dim.
Learning rate placeholder with rank 0.

Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    # TODO: Implement Function
    inputs_real = tf.placeholder(tf.float32, (None, image_width, image_height, image_channels), name='inputs_real')
    inputs_z    = tf.placeholder(tf.float32, (None, z_dim),name='inputs_z')
    learning_rate = tf.placeholder(tf.float32, name='learning_rate')
    return inputs_real, inputs_z, learning_rate


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

Discriminator¶

Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the discriminator, tensor logits of the discriminator).

def discriminator(images, reuse=False):
    """
    Create the discriminator network
    :param images: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    # TODO: Implement Function
    with tf.variable_scope( 'discriminator', reuse=reuse):
        alpha = 0.2
        # Input layer is 28x28x?
        x1 = tf.layers.conv2d(images, 64, 5, strides=2, padding='same')
        x1 = tf.maximum(alpha * x1, x1)
        # 14x14x64
        
        x2 = tf.layers.conv2d(x1, 128, 5, strides=2, padding='same')
        x2 = tf.layers.batch_normalization(x2, training=True)
        x2 = tf.maximum(alpha * x2, x2)
        # 7x7x128
        
        x3 = tf.layers.conv2d(x2, 256, 5, strides=2, padding='same')
        x3 = tf.layers.batch_normalization(x3, training=True)
        relu3 = tf.maximum(alpha * x3, x3)
        # 4x4x256

        # Flatten it
        flat = tf.reshape(x3, (-1, 4*4*256))
        logits = tf.layers.dense(flat, 1)
        out = tf.sigmoid(logits)
        
        return out, logits


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

Generator¶

Implement generator to generate an image using z. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.

def generator(z, out_channel_dim, is_train=True):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    # TODO: Implement Function
    with tf.variable_scope('generator', reuse=not is_train):
        alpha = 0.2
        # First fully connected layer
        x1 = tf.layers.dense(z, 7*7*512)
        # Reshape it to start the convolutional stack
        x1 = tf.reshape(x1, (-1, 7, 7, 512))
        x1 = tf.layers.batch_normalization(x1, training=is_train)
        x1 = tf.maximum(alpha * x1, x1)
        # 7x7x512 now
        
        x2 = tf.layers.conv2d_transpose(x1, 256, 5, strides=1, padding='same')
        x2 = tf.layers.batch_normalization(x2, training=is_train)
        x2 = tf.maximum(alpha * x2, x2)
        # 7x7x256 now
        
        x3 = tf.layers.conv2d_transpose(x2, 128, 5, strides=2, padding='same')
        x3 = tf.layers.batch_normalization(x3, training=is_train)
        x3 = tf.maximum(alpha * x3, x3)
        # 14x14x128 now
        
        # Output layer
        logits = tf.layers.conv2d_transpose(x3, out_channel_dim, 5, strides=2, padding='same')
        # 28x28xout_channel_dim now
        
        out = tf.tanh(logits)
        
        return out

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

Loss¶

Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    # TODO: Implement Function
    alpha = 0.9
    g_model = generator(input_z, out_channel_dim)
    d_model_real, d_logits_real = discriminator(input_real)
    d_model_fake, d_logits_fake = discriminator(g_model, reuse=True)

    d_loss_real = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real, labels=tf.ones_like(d_logits_real) * alpha))
    d_loss_fake = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.zeros_like(d_model_fake)))
    g_loss = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.ones_like(d_model_fake)))

    d_loss = d_loss_real + d_loss_fake

    return d_loss, g_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

Optimization¶

Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """   
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)): 
        t_vars = tf.trainable_variables()
        d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
        g_vars = [var for var in t_vars if var.name.startswith('generator')]
        # Optimize
        d_train_opt = tf.train.AdamOptimizer(learning_rate=learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
        g_train_opt = tf.train.AdamOptimizer(learning_rate=learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)

    return d_train_opt, g_train_opt


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

Neural Network Training¶

Show Output¶

Use this function to show the current output of the generator during training. It will help you determine how well the GANs is training.

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

Train¶

Implement train to build and train the GANs. Use the following functions you implemented:

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    # TODO: Build Model
    n_images = data_shape[0]
    image_width = data_shape[1]
    image_height = data_shape[2]
    image_channels = data_shape[3]
    
    ### Build GAN
    
    # Define Model_Inputs
    input_real, input_z, lr = model_inputs(image_width, image_height, image_channels, z_dim)
    
    # Define Model_Loss
    d_loss, g_loss = model_loss(input_real, input_z, image_channels)
    
    # Define Model_Opt
    d_train_opt, g_train_opt = model_opt(d_loss, g_loss, lr, beta1)
    
    ### Training
    n_total_batches = n_images // batch_size
    
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            nbSteps = 0
            for batch_images in get_batches(batch_size):
                nbSteps += 1

                # Scale real image values from [-0.5, 0.5] to [-1, 1]
                batch_images = batch_images*2.0
                
                # Sample random noise for generator
                batch_z = np.random.uniform(-1, 1, size=(batch_size, z_dim))
                
                # Run optimizers
                # run discriminator optimizer
                _ = sess.run(d_train_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate}) 
                # run generator optimizer
                _ = sess.run(g_train_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
                
                # For each 10 batches, get the losses and print them out                
                if nbSteps % 10 == 0:
                    d_train_loss = d_loss.eval({input_real: batch_images, input_z: batch_z})                    
                    g_train_loss = g_loss.eval({input_z: batch_z})
                    
                    print("Epoch {}/{} , Batch {}/{}: ".format(epoch_i+1, epoch_count, nbSteps, n_total_batches),
                         "Discriminator loss = {:.4f} ".format(d_train_loss),
                         "Generator loss = {:.4f}".format(g_train_loss))
                    
                # Show generator output samples so we can see the progress during training
                
                if nbSteps % 100 == 0:
                    show_generator_output(sess, show_n_images, input_z, image_channels, data_image_mode)

MNIST¶

Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.

batch_size = 32
z_dim = 128
learning_rate = 0.001
beta1 = 0.5

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Epoch 1/2 , Batch 10/1875:  Discriminator loss = 0.6969  Generator loss = 9.5488
Epoch 1/2 , Batch 20/1875:  Discriminator loss = 2.9969  Generator loss = 10.8300
Epoch 1/2 , Batch 30/1875:  Discriminator loss = 2.2273  Generator loss = 11.0765
Epoch 1/2 , Batch 40/1875:  Discriminator loss = 0.6439  Generator loss = 2.5999
Epoch 1/2 , Batch 50/1875:  Discriminator loss = 2.9662  Generator loss = 0.1843
Epoch 1/2 , Batch 60/1875:  Discriminator loss = 1.8691  Generator loss = 0.6017
Epoch 1/2 , Batch 70/1875:  Discriminator loss = 2.3962  Generator loss = 3.5931
Epoch 1/2 , Batch 80/1875:  Discriminator loss = 1.3759  Generator loss = 3.9895
Epoch 1/2 , Batch 90/1875:  Discriminator loss = 2.2891  Generator loss = 4.1657
Epoch 1/2 , Batch 100/1875:  Discriminator loss = 1.0149  Generator loss = 2.5679

Epoch 1/2 , Batch 110/1875:  Discriminator loss = 1.0501  Generator loss = 1.1543
Epoch 1/2 , Batch 120/1875:  Discriminator loss = 0.9937  Generator loss = 1.4242
Epoch 1/2 , Batch 130/1875:  Discriminator loss = 1.2648  Generator loss = 0.7349
Epoch 1/2 , Batch 140/1875:  Discriminator loss = 1.5119  Generator loss = 3.1196
Epoch 1/2 , Batch 150/1875:  Discriminator loss = 1.3064  Generator loss = 1.0227
Epoch 1/2 , Batch 160/1875:  Discriminator loss = 1.7587  Generator loss = 0.3863
Epoch 1/2 , Batch 170/1875:  Discriminator loss = 1.5440  Generator loss = 1.2082
Epoch 1/2 , Batch 180/1875:  Discriminator loss = 1.5004  Generator loss = 1.0571
Epoch 1/2 , Batch 190/1875:  Discriminator loss = 2.1303  Generator loss = 2.1605
Epoch 1/2 , Batch 200/1875:  Discriminator loss = 1.8213  Generator loss = 0.3740

Epoch 1/2 , Batch 210/1875:  Discriminator loss = 1.0369  Generator loss = 0.9579
Epoch 1/2 , Batch 220/1875:  Discriminator loss = 1.6412  Generator loss = 0.6856
Epoch 1/2 , Batch 230/1875:  Discriminator loss = 1.5897  Generator loss = 1.7077
Epoch 1/2 , Batch 240/1875:  Discriminator loss = 1.1474  Generator loss = 1.0221
Epoch 1/2 , Batch 250/1875:  Discriminator loss = 1.3049  Generator loss = 0.7055
Epoch 1/2 , Batch 260/1875:  Discriminator loss = 1.2522  Generator loss = 0.9566
Epoch 1/2 , Batch 270/1875:  Discriminator loss = 1.7855  Generator loss = 0.3331
Epoch 1/2 , Batch 280/1875:  Discriminator loss = 1.5878  Generator loss = 0.4720
Epoch 1/2 , Batch 290/1875:  Discriminator loss = 1.3879  Generator loss = 0.7600
Epoch 1/2 , Batch 300/1875:  Discriminator loss = 1.3392  Generator loss = 0.8874

Epoch 1/2 , Batch 310/1875:  Discriminator loss = 1.5493  Generator loss = 0.4314
Epoch 1/2 , Batch 320/1875:  Discriminator loss = 1.8788  Generator loss = 0.3707
Epoch 1/2 , Batch 330/1875:  Discriminator loss = 1.5570  Generator loss = 0.4958
Epoch 1/2 , Batch 340/1875:  Discriminator loss = 1.2658  Generator loss = 0.9464
Epoch 1/2 , Batch 350/1875:  Discriminator loss = 1.4464  Generator loss = 1.0479
Epoch 1/2 , Batch 360/1875:  Discriminator loss = 1.7002  Generator loss = 0.9331
Epoch 1/2 , Batch 370/1875:  Discriminator loss = 1.6380  Generator loss = 0.4259
Epoch 1/2 , Batch 380/1875:  Discriminator loss = 1.2910  Generator loss = 1.8386
Epoch 1/2 , Batch 390/1875:  Discriminator loss = 1.1547  Generator loss = 1.0941
Epoch 1/2 , Batch 400/1875:  Discriminator loss = 1.2901  Generator loss = 0.9931

Epoch 1/2 , Batch 410/1875:  Discriminator loss = 1.2723  Generator loss = 0.7278
Epoch 1/2 , Batch 420/1875:  Discriminator loss = 1.3599  Generator loss = 0.6483
Epoch 1/2 , Batch 430/1875:  Discriminator loss = 1.1117  Generator loss = 0.8610
Epoch 1/2 , Batch 440/1875:  Discriminator loss = 1.4146  Generator loss = 1.2763
Epoch 1/2 , Batch 450/1875:  Discriminator loss = 1.7647  Generator loss = 0.3755
Epoch 1/2 , Batch 460/1875:  Discriminator loss = 1.8651  Generator loss = 0.3279
Epoch 1/2 , Batch 470/1875:  Discriminator loss = 1.4220  Generator loss = 0.6717
Epoch 1/2 , Batch 480/1875:  Discriminator loss = 1.3182  Generator loss = 0.6295
Epoch 1/2 , Batch 490/1875:  Discriminator loss = 1.3186  Generator loss = 0.5931
Epoch 1/2 , Batch 500/1875:  Discriminator loss = 1.3420  Generator loss = 0.8371

Epoch 1/2 , Batch 510/1875:  Discriminator loss = 1.7099  Generator loss = 0.4068
Epoch 1/2 , Batch 520/1875:  Discriminator loss = 1.3314  Generator loss = 0.6804
Epoch 1/2 , Batch 530/1875:  Discriminator loss = 1.6986  Generator loss = 0.4358
Epoch 1/2 , Batch 540/1875:  Discriminator loss = 1.3732  Generator loss = 1.0771
Epoch 1/2 , Batch 550/1875:  Discriminator loss = 1.4061  Generator loss = 0.6797
Epoch 1/2 , Batch 560/1875:  Discriminator loss = 1.4326  Generator loss = 0.5797
Epoch 1/2 , Batch 570/1875:  Discriminator loss = 1.3732  Generator loss = 0.6959
Epoch 1/2 , Batch 580/1875:  Discriminator loss = 1.3136  Generator loss = 0.9127
Epoch 1/2 , Batch 590/1875:  Discriminator loss = 1.5025  Generator loss = 1.0456
Epoch 1/2 , Batch 600/1875:  Discriminator loss = 1.1984  Generator loss = 0.7370

Epoch 1/2 , Batch 610/1875:  Discriminator loss = 1.7184  Generator loss = 0.3873
Epoch 1/2 , Batch 620/1875:  Discriminator loss = 1.5278  Generator loss = 0.9575
Epoch 1/2 , Batch 630/1875:  Discriminator loss = 1.2713  Generator loss = 1.0722
Epoch 1/2 , Batch 640/1875:  Discriminator loss = 1.1795  Generator loss = 0.7857
Epoch 1/2 , Batch 650/1875:  Discriminator loss = 1.4065  Generator loss = 1.0290
Epoch 1/2 , Batch 660/1875:  Discriminator loss = 1.4036  Generator loss = 0.8619
Epoch 1/2 , Batch 670/1875:  Discriminator loss = 1.3705  Generator loss = 0.8485
Epoch 1/2 , Batch 680/1875:  Discriminator loss = 1.2010  Generator loss = 0.8289
Epoch 1/2 , Batch 690/1875:  Discriminator loss = 1.4602  Generator loss = 0.7898
Epoch 1/2 , Batch 700/1875:  Discriminator loss = 1.2750  Generator loss = 1.0715

Epoch 1/2 , Batch 710/1875:  Discriminator loss = 1.5522  Generator loss = 0.4395
Epoch 1/2 , Batch 720/1875:  Discriminator loss = 1.4960  Generator loss = 0.5035
Epoch 1/2 , Batch 730/1875:  Discriminator loss = 1.5864  Generator loss = 1.5463
Epoch 1/2 , Batch 740/1875:  Discriminator loss = 1.7872  Generator loss = 0.3769
Epoch 1/2 , Batch 750/1875:  Discriminator loss = 1.2022  Generator loss = 1.1784
Epoch 1/2 , Batch 760/1875:  Discriminator loss = 1.3194  Generator loss = 1.2084
Epoch 1/2 , Batch 770/1875:  Discriminator loss = 1.5119  Generator loss = 0.7935
Epoch 1/2 , Batch 780/1875:  Discriminator loss = 1.3220  Generator loss = 0.7279
Epoch 1/2 , Batch 790/1875:  Discriminator loss = 1.3880  Generator loss = 0.6686
Epoch 1/2 , Batch 800/1875:  Discriminator loss = 1.5516  Generator loss = 0.4435

Epoch 1/2 , Batch 810/1875:  Discriminator loss = 1.6529  Generator loss = 0.3852
Epoch 1/2 , Batch 820/1875:  Discriminator loss = 1.3807  Generator loss = 0.6401
Epoch 1/2 , Batch 830/1875:  Discriminator loss = 1.4600  Generator loss = 1.4773
Epoch 1/2 , Batch 840/1875:  Discriminator loss = 1.8502  Generator loss = 0.3082
Epoch 1/2 , Batch 850/1875:  Discriminator loss = 1.8546  Generator loss = 0.3025
Epoch 1/2 , Batch 860/1875:  Discriminator loss = 1.2707  Generator loss = 0.8900
Epoch 1/2 , Batch 870/1875:  Discriminator loss = 1.4337  Generator loss = 1.8082
Epoch 1/2 , Batch 880/1875:  Discriminator loss = 1.3596  Generator loss = 1.0335
Epoch 1/2 , Batch 890/1875:  Discriminator loss = 1.2740  Generator loss = 1.1512
Epoch 1/2 , Batch 900/1875:  Discriminator loss = 1.3192  Generator loss = 0.6657

Epoch 1/2 , Batch 910/1875:  Discriminator loss = 1.5203  Generator loss = 1.2402
Epoch 1/2 , Batch 920/1875:  Discriminator loss = 1.4589  Generator loss = 0.4792
Epoch 1/2 , Batch 930/1875:  Discriminator loss = 1.6910  Generator loss = 1.8771
Epoch 1/2 , Batch 940/1875:  Discriminator loss = 1.3430  Generator loss = 0.8473
Epoch 1/2 , Batch 950/1875:  Discriminator loss = 1.4848  Generator loss = 1.6144
Epoch 1/2 , Batch 960/1875:  Discriminator loss = 1.3914  Generator loss = 0.6617
Epoch 1/2 , Batch 970/1875:  Discriminator loss = 1.2760  Generator loss = 0.9061
Epoch 1/2 , Batch 980/1875:  Discriminator loss = 1.3654  Generator loss = 1.4552
Epoch 1/2 , Batch 990/1875:  Discriminator loss = 1.2739  Generator loss = 0.7228
Epoch 1/2 , Batch 1000/1875:  Discriminator loss = 1.4024  Generator loss = 1.0874

CelebA¶

Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.

batch_size = 64
z_dim = 100
learning_rate = 0.0001
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 1

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

Epoch 1/1 , Batch 10/3165:  Discriminator loss = 3.2419  Generator loss = 0.2056
Epoch 1/1 , Batch 20/3165:  Discriminator loss = 1.7079  Generator loss = 0.7680
Epoch 1/1 , Batch 30/3165:  Discriminator loss = 1.2746  Generator loss = 1.0463
Epoch 1/1 , Batch 40/3165:  Discriminator loss = 0.9965  Generator loss = 1.7740
Epoch 1/1 , Batch 50/3165:  Discriminator loss = 1.0335  Generator loss = 1.1640
Epoch 1/1 , Batch 60/3165:  Discriminator loss = 0.7775  Generator loss = 2.0621
Epoch 1/1 , Batch 70/3165:  Discriminator loss = 1.0371  Generator loss = 1.0416
Epoch 1/1 , Batch 80/3165:  Discriminator loss = 1.1398  Generator loss = 3.2216
Epoch 1/1 , Batch 90/3165:  Discriminator loss = 0.7707  Generator loss = 1.8309
Epoch 1/1 , Batch 100/3165:  Discriminator loss = 1.3335  Generator loss = 1.1310

Epoch 1/1 , Batch 110/3165:  Discriminator loss = 1.1735  Generator loss = 1.8474
Epoch 1/1 , Batch 120/3165:  Discriminator loss = 1.1538  Generator loss = 1.0542
Epoch 1/1 , Batch 130/3165:  Discriminator loss = 1.1648  Generator loss = 1.2348
Epoch 1/1 , Batch 140/3165:  Discriminator loss = 1.1673  Generator loss = 0.7439
Epoch 1/1 , Batch 150/3165:  Discriminator loss = 1.6224  Generator loss = 4.0203
Epoch 1/1 , Batch 160/3165:  Discriminator loss = 0.8522  Generator loss = 1.4974
Epoch 1/1 , Batch 170/3165:  Discriminator loss = 0.8830  Generator loss = 1.3580
Epoch 1/1 , Batch 180/3165:  Discriminator loss = 1.1569  Generator loss = 0.9760
Epoch 1/1 , Batch 190/3165:  Discriminator loss = 1.5442  Generator loss = 0.6247
Epoch 1/1 , Batch 200/3165:  Discriminator loss = 1.1457  Generator loss = 1.0386

Epoch 1/1 , Batch 210/3165:  Discriminator loss = 1.0095  Generator loss = 1.5542
Epoch 1/1 , Batch 220/3165:  Discriminator loss = 1.1394  Generator loss = 0.9018
Epoch 1/1 , Batch 230/3165:  Discriminator loss = 1.3987  Generator loss = 0.8097
Epoch 1/1 , Batch 240/3165:  Discriminator loss = 1.4179  Generator loss = 0.5377
Epoch 1/1 , Batch 250/3165:  Discriminator loss = 1.2544  Generator loss = 0.8631
Epoch 1/1 , Batch 260/3165:  Discriminator loss = 1.2239  Generator loss = 0.8455
Epoch 1/1 , Batch 270/3165:  Discriminator loss = 1.0970  Generator loss = 1.0345
Epoch 1/1 , Batch 280/3165:  Discriminator loss = 1.1882  Generator loss = 0.8610
Epoch 1/1 , Batch 290/3165:  Discriminator loss = 1.1811  Generator loss = 1.3814
Epoch 1/1 , Batch 300/3165:  Discriminator loss = 1.3964  Generator loss = 0.8443

Epoch 1/1 , Batch 310/3165:  Discriminator loss = 1.1952  Generator loss = 1.2088
Epoch 1/1 , Batch 320/3165:  Discriminator loss = 1.3946  Generator loss = 1.2641
Epoch 1/1 , Batch 330/3165:  Discriminator loss = 1.2215  Generator loss = 1.3173
Epoch 1/1 , Batch 340/3165:  Discriminator loss = 1.4621  Generator loss = 0.6847
Epoch 1/1 , Batch 350/3165:  Discriminator loss = 1.3106  Generator loss = 0.8224
Epoch 1/1 , Batch 360/3165:  Discriminator loss = 1.1159  Generator loss = 1.0730
Epoch 1/1 , Batch 370/3165:  Discriminator loss = 1.1213  Generator loss = 0.7788
Epoch 1/1 , Batch 380/3165:  Discriminator loss = 1.1500  Generator loss = 1.3363
Epoch 1/1 , Batch 390/3165:  Discriminator loss = 1.1794  Generator loss = 1.0003
Epoch 1/1 , Batch 400/3165:  Discriminator loss = 1.1593  Generator loss = 0.9601

Epoch 1/1 , Batch 410/3165:  Discriminator loss = 1.1475  Generator loss = 1.7013
Epoch 1/1 , Batch 420/3165:  Discriminator loss = 1.4998  Generator loss = 0.5472
Epoch 1/1 , Batch 430/3165:  Discriminator loss = 1.1841  Generator loss = 1.0882
Epoch 1/1 , Batch 440/3165:  Discriminator loss = 1.2385  Generator loss = 0.7230
Epoch 1/1 , Batch 450/3165:  Discriminator loss = 1.1443  Generator loss = 1.2468
Epoch 1/1 , Batch 460/3165:  Discriminator loss = 1.4880  Generator loss = 0.9434
Epoch 1/1 , Batch 470/3165:  Discriminator loss = 1.6229  Generator loss = 0.7483
Epoch 1/1 , Batch 480/3165:  Discriminator loss = 1.8353  Generator loss = 0.4340
Epoch 1/1 , Batch 490/3165:  Discriminator loss = 1.4735  Generator loss = 0.6079
Epoch 1/1 , Batch 500/3165:  Discriminator loss = 1.4855  Generator loss = 0.7190

Epoch 1/1 , Batch 510/3165:  Discriminator loss = 1.7838  Generator loss = 0.9019
Epoch 1/1 , Batch 520/3165:  Discriminator loss = 1.2942  Generator loss = 0.6783
Epoch 1/1 , Batch 530/3165:  Discriminator loss = 1.7561  Generator loss = 0.4148
Epoch 1/1 , Batch 540/3165:  Discriminator loss = 0.9730  Generator loss = 1.9785
Epoch 1/1 , Batch 550/3165:  Discriminator loss = 1.3047  Generator loss = 0.8291
Epoch 1/1 , Batch 560/3165:  Discriminator loss = 1.5347  Generator loss = 0.7343
Epoch 1/1 , Batch 570/3165:  Discriminator loss = 1.3411  Generator loss = 1.0060
Epoch 1/1 , Batch 580/3165:  Discriminator loss = 1.6498  Generator loss = 0.7860
Epoch 1/1 , Batch 590/3165:  Discriminator loss = 1.2737  Generator loss = 0.8204
Epoch 1/1 , Batch 600/3165:  Discriminator loss = 1.5064  Generator loss = 0.7442

Epoch 1/1 , Batch 610/3165:  Discriminator loss = 1.2803  Generator loss = 0.8875
Epoch 1/1 , Batch 620/3165:  Discriminator loss = 1.3925  Generator loss = 0.9389
Epoch 1/1 , Batch 630/3165:  Discriminator loss = 1.3543  Generator loss = 1.0834
Epoch 1/1 , Batch 640/3165:  Discriminator loss = 1.2869  Generator loss = 0.8380
Epoch 1/1 , Batch 650/3165:  Discriminator loss = 1.3962  Generator loss = 1.2198
Epoch 1/1 , Batch 660/3165:  Discriminator loss = 1.5434  Generator loss = 0.5990
Epoch 1/1 , Batch 670/3165:  Discriminator loss = 1.4252  Generator loss = 0.6326
Epoch 1/1 , Batch 680/3165:  Discriminator loss = 1.4810  Generator loss = 0.7571
Epoch 1/1 , Batch 690/3165:  Discriminator loss = 1.5406  Generator loss = 0.9291
Epoch 1/1 , Batch 700/3165:  Discriminator loss = 1.3757  Generator loss = 1.1387

Epoch 1/1 , Batch 710/3165:  Discriminator loss = 1.3651  Generator loss = 0.7184
Epoch 1/1 , Batch 720/3165:  Discriminator loss = 1.2612  Generator loss = 1.0961
Epoch 1/1 , Batch 730/3165:  Discriminator loss = 1.5102  Generator loss = 0.6590
Epoch 1/1 , Batch 740/3165:  Discriminator loss = 1.4399  Generator loss = 0.7731
Epoch 1/1 , Batch 750/3165:  Discriminator loss = 1.4429  Generator loss = 0.7310
Epoch 1/1 , Batch 760/3165:  Discriminator loss = 1.4194  Generator loss = 0.6937
Epoch 1/1 , Batch 770/3165:  Discriminator loss = 1.5084  Generator loss = 0.9431
Epoch 1/1 , Batch 780/3165:  Discriminator loss = 1.2311  Generator loss = 0.9124
Epoch 1/1 , Batch 790/3165:  Discriminator loss = 1.3774  Generator loss = 0.6876
Epoch 1/1 , Batch 800/3165:  Discriminator loss = 1.3203  Generator loss = 0.9138

Epoch 1/1 , Batch 810/3165:  Discriminator loss = 1.4414  Generator loss = 0.9120
Epoch 1/1 , Batch 820/3165:  Discriminator loss = 1.5884  Generator loss = 0.4849
Epoch 1/1 , Batch 830/3165:  Discriminator loss = 1.5773  Generator loss = 0.5532
Epoch 1/1 , Batch 840/3165:  Discriminator loss = 1.2807  Generator loss = 0.9734
Epoch 1/1 , Batch 850/3165:  Discriminator loss = 1.3270  Generator loss = 1.0062
Epoch 1/1 , Batch 860/3165:  Discriminator loss = 1.3425  Generator loss = 0.8544
Epoch 1/1 , Batch 870/3165:  Discriminator loss = 1.4467  Generator loss = 0.6459
Epoch 1/1 , Batch 880/3165:  Discriminator loss = 1.3785  Generator loss = 0.9198
Epoch 1/1 , Batch 890/3165:  Discriminator loss = 1.3593  Generator loss = 0.8061
Epoch 1/1 , Batch 900/3165:  Discriminator loss = 1.2606  Generator loss = 0.9253

Epoch 1/1 , Batch 910/3165:  Discriminator loss = 1.4226  Generator loss = 0.6482
Epoch 1/1 , Batch 920/3165:  Discriminator loss = 1.4399  Generator loss = 0.8591
Epoch 1/1 , Batch 930/3165:  Discriminator loss = 1.3338  Generator loss = 0.7335
Epoch 1/1 , Batch 940/3165:  Discriminator loss = 1.2240  Generator loss = 0.8563
Epoch 1/1 , Batch 950/3165:  Discriminator loss = 1.3452  Generator loss = 0.9326
Epoch 1/1 , Batch 960/3165:  Discriminator loss = 1.4790  Generator loss = 0.7743
Epoch 1/1 , Batch 970/3165:  Discriminator loss = 1.5573  Generator loss = 0.7189
Epoch 1/1 , Batch 980/3165:  Discriminator loss = 1.3631  Generator loss = 0.8377
Epoch 1/1 , Batch 990/3165:  Discriminator loss = 1.2732  Generator loss = 1.1480
Epoch 1/1 , Batch 1000/3165:  Discriminator loss = 1.3253  Generator loss = 0.9563

Submitting This Project¶

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.